Dual Noise-Suppressed ZNN with Predefined-Time Convergence and its Application in Matrix Inversion

Abstract

Original zeroing neural network (OZNN) can effectively solve the problem of matrix inversion. Generally, the problem of matrix inversion is solved in the noiseless environment. However, noises are common, OZNN can not solve the problem with harmonic noise interference. Therefore, the integrated enhanced zeroing neural network (IEZNN) is proposed to overcome this difficulty. IEZNN can deal with the harmonic noise interference problem when the time change slightly. But in the case of large amplitude or frequency, IEZNN has not strong ability to tolerate the noise and the convergence speed is relatively slow. Therefore, by adding a novel nonlinear activation for IEZNN, which also has the ability to suppress noise, a dual noise-suppressed ZNN (DNSZNN) is proposed to solve this problem. DNSZNN not only has good noise suppression characteristics, but also can converge in the predefined time. Finally, the experimental results demonstrate that the DNSZNN has the best robustness and convergence performance under the same external harmonic noise interference compared with the OZNN and the IEZNN.